A Lagrangian analysis of cold cloud clusters and …...Esmaili et al., 2016 1 1 A Lagrangian analysis of cold cloud clusters and their life cycles with 2 satellite observations 3 4

Esmailietal.,2016 1

A Lagrangian analysis of cold cloud clusters and their life cycles with 1

satellite observations 2

3

Rebekah Bradley Esmaili 4

Dept. of Atmospheric and Oceanic Science, University of Maryland, College Park, Maryland 5Yudong Tian 6

Earth System Science Interdisciplinary Center, University of Maryland, College Park, Maryland 7Daniel Alejandro Vila 8

National Institute for Space Research (INPE), São José dos Campos, Brazil 9Kyu-Myong Kim 10

NASA Goddard Space Flight Center, Greenbelt, Maryland 11

12 13

Submitted to Journal of Geophysical Research: Atmospheres 14July 15, 2016 15

16

17Corresponding author address: 18

Rebekah B. Esmaili 193417 Computer & Space Sciences Blg. 20College Park, Maryland 20742-2425 21

Tel: (301) 614-6537 22Email: [email protected]

https://ntrs.nasa.gov/search.jsp?R=20170003683 2020-05-12T10:37:41+00:00Z

Esmailietal.,2016 2

ABSTRACT 24

Cloud movement and evolution signify the complex water and energy transport in the 25

atmosphere-ocean-land system. Detecting, clustering, and tracking clouds as semi-26

coherent cluster objects enables study of their evolution which can complement climate 27

model simulations and enhance satellite retrieval algorithms, where there are large gaps 28

between overpasses. Using an area-overlap cluster tracking algorithm, in this study we 29

examine the trajectories, horizontal extent, and brightness temperature variations of 30

millions of individual cloud clusters over their lifespan, from infrared satellite 31

observations at 30-minute, 4-km resolution, for a period of 11 years. We found that the 32

majority of cold clouds were both small and short-lived and that their frequency and 33

location are influenced by El Niño. More importantly, this large sample of individually 34

tracked clouds shows their horizontal size and temperature evolution. Longer lived 35

clusters tended to achieve their temperature and size maturity milestones at different 36

times, while these stages often occurred simultaneously in shorter lived clusters. On 37

average, clusters with this lag also exhibited a greater rainfall contribution than those 38

where minimum temperature and maximum size stages occurred simultaneously. 39

Furthermore, by examining the diurnal cycle of cluster development over Africa and the 40

Indian subcontinent, we observed differences in the local timing of the maximum 41

occurrence at different life cycle stages. Over land there was a strong diurnal peak in the 42

afternoon while over the ocean there was a semi-diurnal peak composed of longer-lived 43

clusters in the early morning hours and shorter-lived clusters in the afternoon. Building 44

on regional specific work, this study provides a long-term, high-resolution, and global 45

survey of object-based cloud characteristics. 46

47

Esmailietal.,2016 3

1. Introduction 48

Clouds are the most visible vital sign of the atmosphere’s dynamic water and energy 49

transfer. They are responsible for the latent heat release that drives the atmospheric 50

circulation. Their transport of water in the form of moisture and precipitation is critical 51

for the Earth’s hydrological cycle. On sub-synoptic scales, the cloud systems’ movement, 52

evolution, and spatial and temporal characteristics are remarkably turbulent and complex. 53

The Lagrangian framework is an effective approach to study cloud clusters. Treating each 54

cloud as an object across it’s lifespan produces useful information on the evolution of 55

cloud systems’ properties, which is not available from the Eulerian view [Machado et al., 56

1998]. Renewed interest in cloud object-based evolution is partly due to the advancement 57

of satellite-based multi-sensor high-resolution precipitation estimates [Li et al., 2015]. 58

Currently, these estimates rely heavily on observations from passive microwave (PMW) 59

sensors aboard polar-orbiting satellites [Kummerow et al., 2000; Joyce et al., 2004; 60

Huffman et al. 2007; Huffman et al. 2013]. These PMW-based estimates are relatively 61

accurate, but they do not correlate well with surface observations when precipitation is 62

very light, very heavy, or over saturated land surfaces, particularly during winter months 63

[Ebert et al., 2007]. Additionally, these data have large spatial and temporal gaps. 64

One way to bridge these coverage gaps is to use cloud system advection information 65

derived from high-resolution infrared observations to continuously “morph” the PMW-66

based rainfall [Joyce et al., 2004; Kubota et al., 2007; Xie and Xiong, 2011]. These 67

approaches have been proven effective and are being incorporated into Integrated Multi-68

satellitE Retrievals for GPM [IMERG; Huffman et al., 2013], the next-generation, Global 69

Esmailietal.,2016 4

Precipitation Measurement (GPM; Hou et al., 2014) era product suite. However, the 70

accuracy of the PMW-based estimates is also influenced by the life cycle stage [Tadesse 71

and Agnastou, 2011]. Developing a more detailed understanding of evolution can provide 72

additional context. 73

Another critical application is the evaluation and diagnosis of global models. Currently 74

atmospheric models still unrealistically reproduce observed precipitation [Ebert et al., 75

2007; Stephens et al. 2010]. The conventional Eulerian validation gives the spatial and 76

temporal statistics on each individual grid box, which is the accumulation of many 77

different cloud systems at various life stages. A Lagrangian comparison of modeled and 78

observed cloud evolution statistics could produce additional insight on the modeling of 79

individual cloud-precipitation processes [Boer and Ramanathan, 1997]. 80

In the past, studies that combined infrared satellite-based cloud cluster tracking with 81

Tropical Ocean Global Atmosphere (TOGA) field campaigns to examine cloud evolution, 82

anatomy, and development conditions [Williams and Houze, 1987; Chen and Houze, 83

1997]. On larger scales, Mesoscale Convective Systems (MCS) have in particular been 84

studied due to their ease of detection in radar and satellite images and destructiveness 85

[Maddox, 1980; Laing and Fritsch, 1997; Blamey and Reason, 2011]. MCS display 86

regularity in their life cycles, enabling Machado et al. [2004] to develop a statistical life 87

cycle model to predict MCS propagation with good forecasting skill. 88

As global, long-term, quality controlled IR data and precipitation data become available 89

[e.g., Janowiak et al., 2001; Joyce et al., 2004], it becomes feasible to extend IR-based 90

cloud tracking beyond regional scales and expand the scales of observed phenomena. By 91

Esmailietal.,2016 5

following a large number of cloud clusters on the global scale for over 11 years, we will 92

be able to understand more systematically their large-scale dynamical and statistic 93

characteristics. 94

In this paper, we present a near-global (60°S-60°N), high-resolution (30-minute, 4-km), 95

long-term (11-year) study of cloud cluster tracks, life cycle evolution, and diurnal cycle. 96

The high-resolution data used for the study and the methodology for storm tracking are 97

described in Section 2 and 3, respectively. Section 4 presents the results, followed by 98

summary and discussions in Section 5. 99

2. Data 100

For our study, we use the NCEP/CPC a 4-km, half hourly infrared (IR) brightness 101

temperature dataset [Janowiak et al., 2001]. The dataset is merged from all available 102

geostationary satellites (GMS, Meteosat-5, Meteosat-7, GOES-8 and GOES-10) to form 103

near-global (60°N-60°S) coverage on a uniform latitude-longitude grid. We used 11 years 104

of data from 2002 to 2012 for our study. 105

We have performed additional quality control of the IR data. There are gaps in the IR 106

data in regions covered by the GMS satellite in the Western Pacific (120°-170°E), so we 107

filled the missing data by interpolating the preceding and following 30-minute snapshots, 108

to produce more seamless coverage. Our tests show that this interpolation helps to 109

prevent early termination of the cloud lifespan due to missing data. 110

3. Methodology111

Esmailietal.,2016 6

The techniques for tracking clouds are mature and largely similar, albeit there are many 112

implementations. Most of those techniques involve IR geostationary satellite imagery to 113

follow classes of convective events. The primary dissimilarities in algorithms involve the 114

detection criteria, such as through the selection of brightness temperature or size 115

parameter thresholds [Carvalho and Jones, 2001; Morel and Senesi, 2002], usage of more 116

nuanced detection schemes [Lakshmanan, 2009], or the treatment of splits and merges 117

[Fiolleau and Roca, 2013]. 118

In spite of a variety of implementations, Machado et al. [1998] found most of the life 119

cycle statistics are not overly sensitive to the tracking method used. For this paper, we 120

selected Forecast and Tracking the Evolution of Cloud Clusters [ForTrACC; Vila et al., 121

2001] which has a single temperature and system size threshold and merges and splits are 122

treated as special cases for tracking systems (this will be explained in the Section 3.2). 123

ForTrACC’s simplicity enabled us to capture a broader range of cloud species. 124

Tracking clouds involves the following two major steps: 125

3.1. Identification using temperature and morphology 126

Using brightness temperature thresholds to capture clouds has been used in past studies 127

and typically empirically derived to satisfy the research goals [e.g. Blamey and Reason, 128

2001; Velasco and Fritch, 1987; Williams and Houze, 1987]. In general, brightness 129

temperature detection thresholds vary between 235-255 K and tended to be subjectively 130

chosen. However, the cluster areal extent was found to the linearly dependent on cluster 131

threshold and thus not overly sensitive to the exact threshold chosen [Machado et al. 132

1992; Mapes and Houze, 1993]. 133

Esmailietal.,2016 7

To capture a variety of cloud clusters, we used a single 235 K brightness temperature 134

threshold, which in the upper atmosphere corresponds to a height of roughly 10 km (9 135

km) in the tropics (midlatitudes), which is well into the free atmosphere. Additionally, we 136

applied a minimum size threshold of 100 contiguous pixels (1,600 km2 at the equator) at 137

all time steps, thus limiting the study to events at the upper bounds of mesoscale or larger. 138

We excluded smaller scale events because they would be more suitable for regional 139

studies. Figure 1 shows this selection criteria being applied to a typical IR snapshot. A 140

temperature range of 235-245 K have been used in the past to detect cloud clusters (e.g. 141

Williams and Houze, 1987; MapesandHouze,1993;CarvalhoandJones,2001; 142

Machado et al., 2004); the colder threshold is utilized to avoid capturing frozen, high 143

altitude surfaces. The size threshold reduces the number of tracked clouds by filtering out 144

small-scale events and reducing the number of splits and merges. With only a 145

temperature threshold, a single time step can yield over 17,000 cloud clusters. Applying 146

the size threshold decreased to the number to 800-1000 events. 147

3.2. Tracking using area overlap 148

ForTrACC uses an area overlap technique to track the cloud clusters, both forward and 149

backward in time. If two cloud clusters identified in different time steps have any shared 150

pixels, they were considered the same system and assigned a family number. If more than 151

one match was found, the largest overlaping system was tracked. 152

Using infrared data, we show in Figure 2 a schematic of the area-overlap handling in 153

ForTrACC. The area overlap technique produces several cloud cluster merge scenarios: 154

one-to-one (continuous), one-to-many (split), many-to-one (merge), or no match 155

Esmailietal.,2016 8

(initialization or dissipation). Most systems undergo merging or splitting in their life 156

cycle, the prior occurs before maturation, the latter more frequently towards the end of 157

the life cycle. Only one cluster is followed at each time step to keep features well defined. 158

When clusters split, the largest cloud continues to be tracked while the smaller split 159

clusters are treated as a new family and the lifetime clock is reset. All merging clusters 160

are considered a dissipation and their life cycle ends. ForTrACC’s handling of split and 161

merge segments is different from earlier work; in other schemes, the segments remain 162

part of the cloud cluster system rather than considered a new systems [Mapes and Houze, 163

1993; Chen and Houze, 1996; 1997)]. 164

A sample output of the resulting cloud cluster tracks are shown in Figure 3. In addition to 165

showing centroid location, statistics related to the size or areal extent, the mean 166

brightness temperature, and travel distance of the cluster are also calculated. Colder 167

temperatures indicate higher cloud tops while areal extent shows the relative scale of the 168

observed system. All clusters achieve a minimum temperature and maximum size, which 169

we use as criteria for developmental maturity in Section 4.5. We use this information to 170

study the ForTrACC-based cloud clusters’ statistical properties, climatology, life cycle, 171

rainfall contribution, and diurnal cycle. 172

3.3. Collocation of clusters with passive microwave rainfall estimations 173

To examine the rainfall contribution of cloud clusters, we matched PMW precipitation 174

estimates from IMERG [Huffman et al., 2013] with spatially and temporally collocated 175

cloud clusters. 176

Esmailietal.,2016 9

While TRMM-based products have a longer data record, GPM has global coverage so we 177

selected two months of data to examine (June and December 2014). Both datasets were 178

scaled to a common grid (0.1° x 0.1°) and the available rainfall totals were summed for 179

clusters at various life cycle stages. 180

4. Results 181

4.1. Mean trajectories and statistical properties of cloud clusters 182

Tracking on the global scale builds on regional studies and enables us to document many 183

fundamental statistical characteristics of cloud clusters. At any instant, there are on 184

average 800 clusters larger than 1,600 km2 in the Earth’s atmosphere between 60°S and 185

60°N. Figure 4 shows the global distribution of clusters with lifetimes between 6 and 9 186

hours, for both December through February (DJF) and June through August (JJA). The 187

mean trajectories are calculated by averaging the endpoints of all cluster centroids that 188

initiate at the same 2° x 2° binned latitude and longitude coordinates. The colors 189

represent the net zonal direction of the flow. 190

Regarding the zonal average distance travelled by 6-9 hour lifetime clusters in Figure 4, 191

we found that cloud clusters travel further in the Northern Hemisphere during DJF; the 192

average distance traveled peaks at 644.8 km at 36°N, which is likely due to influence of 193

the climatological jet stream on development and propagation. In the Southern 194

Hemisphere the maximum occurs near 52°S at a lower 419.8 km. Movement in the 195

tropics doesn’t vary drastically from each season, but the peak (189.1 km) occurs in JJA 196

at 12°N. This is in part due to the persistence of the ITCZ and African Easterly Wave 197

activity. 198

Esmailietal.,2016 10

Cloud clusters can last from a few hours up to two days, and their sizes range from our 199

minimal threshold to more than 106 km2 (Figure 5). Most of the clusters are short-lived 200

and small (Figure 5a), with 90% of the clusters detected having a size less than 49,275 201

km2 and a lifetime less than 5 hours. The cluster lifetime distribution follows roughly a 202

log-linear distribution while the cluster size distribution appears to be lognormal at 203

certain scales, the latter being consistent with some past findings [Machado et al. 1992; 204

Mapes and Houze, 1993] but different from others [Lovejoy and Schertzer, 2006]. Figure 205

5b shows the kernel density estimate [Rosenblatt, 1956; Parzen, 1962], a non-parametric 206

estimate of the probability density of maximum areal extent of each cluster across several 207

lifetime bins. 208

Overall, Figure 5 shows that the frequency of cluster lifetime and size are proportional. 209

This is similar to the results from Chen et al. [1997], who show a linear correlation 210

between the count of tropical cloud clusters with respect to maximum size and lifetime in 211

the western Pacific. This reinforces that shorter lived events tend to remain small in scale 212

while longer-lived ones achieve greater horizontal scales. 213

These results can be compared with event tracking based on model data [e.g. Bengtsson 214

et al., 2006; Bengtsson et al., 2009; Hoskins and Hodges, 2001; Neu et al., 2013; Sinclair, 215

1994]. Modelling studies typically use vorticity or sea level pressure as the defining 216

feature of midlatitude cyclone storm tracks. Coupled with lower temporal resolution data, 217

this can result in smoother tracks and are larger and longer-lived than the ones shown in 218

Figure 4. The differences are due to tracking definitions but may also be due to the 219

prevalence of lighter rainfall typical in models as compared with observations [Stephens 220

et al., 2010]. 221


4.2. Cloud cluster climatology 222

On the global scale, the clusters exhibit many systematic spatial and temporal 223

characteristics, as seen in the seasonal climatology map of clusters (Figure 6). The map 224

produced is the frequency of clusters at their maximum areal extent for each 2° x 2° 225

latitude-longitude bin. During DJF, the intertropical convergence zone (ITCZ) is closer to 226

the equator and South Pacific convergence zone is intensified. There is increased activity 227

from the midlatitude storm tracks across the North American west coast and Europe. In 228

JJA, tracks capture the northward placement of the ITCZ, Atlantic coastal storms, and the 229

East Asian monsoon. Less activity is found in proximity of the semi-permanent high 230

pressure systems (e.g. Pacific and Bermuda highs in JJA). Artifacts in south Pacific (40°-231

60°S, 120°-160°W) are due to calibration differences between geostationary platforms 232

and the interface of the half-hourly and hourly sampling regions of the Geostationary 233

Meteorological Satellite between 120°-170°E. Note that in Figure 6a, we excluded data 234

from DJF 2006 from 120°-170°E due to intermittent noisy brightness temperature data in 235

this region. 236

The frequency also reveals some regional subtleties in Figure 6b. Over the Southeast-237

Asia islands in the western Pacific Ocean, there are roughly twice as many clusters along 238

coastlines than the surrounding oceanic areas. This region’s combination of topography, 239

land-sea thermal contrast, and available moisture generates storms that are both large in 240

scale and deep, making it it is one of the rainiest places on earth in TRMM-based object 241

studies [Houze et al., 2015]. 242


Interestingly, a high count of clusters does not necessarily correlate with intense rainfall. 243

Outside the ITCZ, the Amazon, the Asian monsoon, and West African monsoon are 244

among the most active continental regions in terms of cluster frequency. However, 245

TRMM-based studies have shown objects tend to be moderate strength and larger scale in 246

the Amazon while the latter two regions are composed of deep convection [Zipser et al. 247

2006; Houze et al., 2015]. In the Amazon, rainfall features have a lower mean height than 248

those over the Asian and African monsoon regions and warm rain tends to be the greatest 249

contributor of rainfall [Liu and Zipser, 2008]. While not shown, statistically we found 250

that clusters in our study were typically larger, colder, and longer-lived over Western 251

Africa and the Indian Subcontinent (JJA), whereas shorter-lived, moderate sized clusters 252

tended to occur over the Amazon (DJF). 253

Compared to results based on reanalysis-based tracking results, the JJA cluster counts 254

shown in Figure 6b resemble vorticity-based African Easterly [Thorncroft and Hodges, 255

2001]. In both studies, intiation maxima occur along the West Africa coast and Ethiopian 256

highlands as well as over the Pacific, downstream of Central America. We visually 257

observed that our IR-based tracks are noisier than reanalysis derived ones and are less 258

exclusive. Tracking with six-hourly data can skew results towards stronger, longer-lived 259

events, and can miss younger events. 260

4.3. Inter-annual variability 261

There is significant inter-annual variability in cluster occurrence, particularly between El 262

Niño and La Niña years. Figure 7 shows the composite of frequency difference of cluster 263

overpasses at their maximum size during the El Niño phases for 11-years of DJF, binned 264


by 2° x 2° latitude-longitude boxes. This was produced by subtracting the annual average 265

frequency of cluster occurrence during warm phases from the annual average of cool 266

phases. Only seasons with weak, moderate, or strong phases based on the NINO3.4 sea 267

surface temperature anomaly index are included. 268

El Niño has an expected effect on the frequency of cloud clusters in the tropics: more 269

clusters are observed near the equator during the warm phase in the central pacific 270

(160°E-160°W) and in the western pacific (110°-160°E) during the cool phase. However, 271

teleconnections can also be observed; there is an increase in occurrence over the 272

Northwest United States (25-55°N, 100-120°W) and Indian Ocean (10°S-10°N, 40°-273

80°E) and a decrease in the Atlantic basin (10°S-10°N, 60°-10°W) during El Niño. Teng 274

et al. [2014] have shown that there are both increases in cloud cluster occurrence as well 275

as their likelihood of forming tropical cyclones in the western North Pacific during El 276

Niño. 277

4.4. Life cycle of cloud clusters 278

The advantage of continuous Lagrangian tracking is that it allows us to examine 279

systematically the clusters’ full life cycle and the associated evolution of their 280

characteristics. Figures 8 and 9 show how the size and brightness temperature of clusters 281

evolve throughout their lifespan. Each curve represents the average of the aggregated 282

clusters that lived to the same age. For clarity, clusters that merged into or split off from 283

existing clusters were not included in Figures 8 and 9. Shorter curves represent brief 284

events while longer lines represent clusters with longer lifespans. The observed mean life 285

cycles have a well-defined stages of development – initial detection, intensification, 286


maturity, and decay. This can be seen in both their size evolution (Figure 8) and 287

brightness temperature evolution (Figure 9). With respect to size, clusters initiate, grow, 288

and achieve their areal maximum closer to the end of their life cycle (Figure 8). At their 289

size maximum, longer-lived clusters can double or triple their initial areal extent. Shorter-290

lived ones undergo rapid decay early in their cycle. In contrast, during their brightness 291

temperature life cycle, clusters cool to a minimum and then begin to warm for the rest of 292

the life cycle (Figure 9). While an individual clusters’ evolution usually appears erratic 293

and unpredictable, collectively their mean behavior computed from the ensemble of 10 294

million clusters shows regularity. 295

The minimum brightness temperature is reached at an earlier point in the clusters life 296

cycle than the size maximum. This could be due to overshooting tops, which reach deep 297

into the troposphere or lower stratosphere first, and then expand to form anvils as they 298

cool, and thus attaining their minimum brightness temperature before their maximum 299

areal extent. Additionally, clusters at their maximum areal extent produce cirrus shields 300

can also conceal the true extent of the clusters underneath. 301

On the global scale, the life cycle evolution shows substantial differences over 302

contrasting seasons and land surfaces. Due to their similarity, in Figures 8 and 9 regions 303

are divided into seasons along the ±25° latitude line, where Northern and Southern 304

winters (summers) are during DJF and JJA (JJA and DJF), respectively. The tropics use 305

data from both seasons. Generally, growth is more vigorous in summer than in winter 306

(e.g., compare Figures 8b and 8f, Figures 9b and 9f), over land than over ocean (e.g., 307

compare Figures 8a and 8b to Figures 9a and 9b). In addition, the wintertime clusters are 308

much larger than summertime (e.g., Figures 8a and 8e). In the summer, the midlatitude 309


size curves (Figures 8a and 8b) are more similar to the tropics (Figures 8c and 8d). 310

Regarding brightness temperature, there is a larger spread during the summer (Figures 9a 311

and 9b) and in tropics (Figures 9c and 9d) than during the winter for both land and ocean 312

(Figures 9e and 9f). Clusters in the tropics (Figures 9c and 9d) are significantly cooler 313

than higher latitudes due to deep convection (Figures 9a and 9b). 314

The peaks in Figures 8 and 9 were fitted to a quadratic linear regression model to show 315

the general trend of size and temperature maturity across different lifetimes. Shorter-lived 316

clusters tended to be already at maturity at the time of detection – that is, the shortest 317

lines in Figures 8 and 9 show that these clusters total area decreased and temperatures 318

rapidly increased. For longer-lived clusters, the timing of the maximum areal extent and 319

minimum temperature was asynchronous and larger than that for shorter-lived events. We 320

will examine some of the implications of this in the following sections. 321

4.5 Cloud Clusters and Rainfall322

In raining cloud clusters, the differences in the timing of the minimum brightness 323

temperature and maximum size contribute varying amounts to total precipitation. In 324

Figure 10, we identified several distinct life cycle stages (initiation, mixed maturity, 325

minimum brightness temperature, maximum size, and dissipation) and the instantaneous 326

total volumetric rainfall that is attributed to each rain rate bin. Using the procedure 327

detailed in Section 3.3, this was determined by collocating the cloud clusters with 328

available microwave-only rainfall estimations from IMERG [Huffman et al., 2013], for 329

June and December 2014. 330


Due to the lower temporal resolution of polar orbiting satellites, most clouds could only 331

be sampled once, so results are examined in a statistical sense rather than as totals by 332

individual objects. Here we define the minimum temperature (maximum size) as the 333

lowest average temperature (largest areal extent) achieved by a clusters. We also divide 334

contribution into two mutually exclusive maturity states, synchronous and asynchronous 335

occurrence of minimum temperature and maximum areal extent. The prior is denoted as 336

mixed maturity, while the latter is broken down into the two stages of its variables. 337

Collectively, the figure shows the rainfall contribution of the beginning, mature, and final 338

life cycle stage.339

Initially, raining clusters are composed of lighter rain and produce less of it. As 340

development continues, they produce larger volumes of rain as the areal extent of the 341

cloud increases. It is interesting that in all cases, mixed maturity clusters contribute less 342

rainfall than those with asynchronous stages. These cases tended to be shorter lived on 343

average (1.9 hours) than those with larger differences in timing (2.9 hours). 344

There are seasonal differences in these values. The winter midlatitudes (Figurs 10b and 345

10e) produced more overall rain than their corresponding summer hemisphere (Figures 346

10a and 10f) and were more heavily skewed towards lighter rainfall. The tropics had less 347

seasonal variation in rainfall contribution (Figures 10c and 10d). 348

Precipitation retrieval algorithms may benefit from incorporating information on the life 349

cycle stage, season, and hemisphere of the IR cloud cluster. In morphing techniques, the 350

shape and intensity of rain clusters is held constant between overpasses [Joyce et al., 351

2004], while in figures 8 and 9 we show that both horizontal size and temperature growth 352


rates are not constant during cloud cluster evolution. Biases in hourly rain volume 353

estimates vary across life cycle stages, lifetimes, and precipitation algorithm [Tadesse 354

and Anagnostou, 2009]. Knowing the age of the cloud could be useful in devising the 355

next-generation multi-sensor algorithms.356

4.6 Diurnal cycle of cluster evolution 357

By continuously tracking cloud clusters, we can study when and where they reach their 358

life cycle milestones. Figure 11 shows the local solar time (LST) of the maximum in the 359

frequency of cluster initiation. This was calculated from frequency maximum at each 360

hourly, 2° x 2° bin for clusters with a lifetime greater than two hours. Over land, peak 361

cloud initiation occurs in the afternoon, especially in the summer hemisphere. Over the 362

ocean, there is greater prevalence of early morning and afternoon clouds, but the timing 363

of peak activity depends on region. This double peak was also previously found in the 364

West Pacific warm pool by Chen and Houze (1997). To examine these differences in 365

context of development stage, we examine two regions centered over West Africa (0-366

40°N, 50°W-20°E) and the South Asian peninsula (0°-30°N and 60°-90°E). 367

In Figure 12, we examine the kernel density of the LST by cluster life cycle stage in these 368

two regions for both seasons. Over land, there is a strong diurnal cycle and a lag in the 369

local timing of initiation, minimum temperature, and maximum size. The timing 370

differences are much smaller over the ocean in both regions and there is a semi-diurnal 371

cycle over the ocean. The timing of peak initiation over land is earlier in the South Asian 372

region (1300 LST) than in the Western Africa region (1500 LST). This is possibly due to 373

the windward side of the Indian subcontinent skewing the the population to lower 374


initiation times. Over the ocean, the early morning peaks have similar timing (0200 LST), 375

but the afternoon peak in Western Africa is earlier (1100 versus 1300 LST). Kikuchi and 376

Wang [2008] observed this semi-diurnal cycle over the Pacific, Indian, and Atlantic 377

Oceans in empirical orthogonal modes of TRMM datasets. We can take advantage of the 378

known lifetime and further inspect the duration of these cloud clusters at different times 379

of the day. 380

In Figure 13, we show the kernel density for the LST grouped by the three-hourly binned 381

lifetime. Over land, the timing differences were delayed by not more than an hour for all 382

lifetime groups. Shorter-lived clusters (those with lifetimes 6 hours or less) had a sharper 383

peak than longer-lived events (those with lifetimes greater than 6 hours). There are trivial 384

differences in the onset of short versus long-lived events in the South Asia region than 385

over West Africa. 386

However, over the ocean, longer-lived clusters had a greater tendency to occur in the 387

early morning hours, peaking between 0300-0400 and 0400-0500 LST in South Asia and 388

West Africa, respectively. Shorter-lived events peaked both in the early morning and 389

afternoon, but were the primary type in the afternoon afternoon between 1200-1300 LST 390

for both regions. In South Asia, the maxima of short-lived clusters precede that of long-391

lived ones by an hour, partly due to rapid growth and decay of isolated convective cells 392

which upon visual inspection are more numerous in this region than in West Africa. 393

These results are interesting in light of previous examination of TRMM precipitation 394

features, which show that nocturnal storms are more intense over the ocean while over 395

land the strongest storms are observed during the day [Zipser, 2006]. In summary, the 396


oceanic semi-diurnal cycle can be understood to be composed of not just two different 397

size classes, but as cloud clusters with differing lifetimes as well.398

Expanding on the study by Chen and Houze (1997), our results show that their results 399

extend beyond the West Pacific to other regions and over longer time periods. Chen and 400

Houze found that large scale, long lived clusters follow a two-day cycle. The formation 401

of long-lived clusters suppresses subsequent development in that area due to dry 402

downdrafts from strong storms and the reduction in sea surface temperatures due to cloud 403

canopy shading. Examining the development-suppression cycle of cloud clusters in other 404

oceanic regions could be an interesting future direction for this work.405

5. Summary and Discussion406

In this study, we tracked cloud clusters on the global scale to study the climatology and 407

life cycles across a broad class of clusters using 11 years of the high-resolution, satellite-408

based globally merged cloud brightness temperature data. We examined the trajectories, 409

climatology, life cycles, and diurnal cycle of clusters in context of their life cycle stage 410

and lifetime. 411

We found that the vast majority of clusters are short lived and small, demonstrating the 412

need to work with high-resolution data to fill in coverage gaps. Differences in the shapes 413

and scales of life cycle curves reflect the variety of clouds captured and show that 414

evolution is a complex process. Development over the oceans is less intense compared to 415

land, where strong thermal contrast, orography, and aerosols can influence evolution. We 416

observe a larger lag in the occurrence of minimum temperature and maximum size for 417

longer-lived cloud clusters, particularly over land. The diurnal cycle of cloud clusters 418


over the South Asia and West Africa revealed a strong diurnal peak over land and a semi-419

diurnal cycle over the ocean, the latter of which showed greater prevalence of shorter 420

lived cloud clusters in the afternoon and dominance of longer lived events in the early 421

morning. 422

The capability for infrared data to reliably identify and track smaller scale convective 423

systems is an aspect in which global climate models still have difficulties [Stephens et al. 424

2010; Westra et al., 2014]. Thus, IR-based cloud tracking can be used to evaluate the 425

effectiveness of the downscaling abilities of models [Boer and Ramanathan, 1997]. On 426

the other hand, the infrared data can only depict the two-dimensional, cloud-top 427

characteristics of the clusters. To address the complex three-dimensional hydro-thermo-428

dynamics of cloud systems, one has to combine observations from other satellites, such 429

as CloudSat and CALIPSO, with reanalysis data. 430

There are several limitations to this study that represent an area of ongoing work, 431

particularly regarding thresholds. Being too selective on size scales can exclude these 432

events; being too relaxed produces too many splits, which prematurely terminates the 433

cluster. Cold surfaces are a particular challenge, such as the Tibetan Plateau which is dry 434

in the northern winter. However, the relatively high frequency over this region in Figure 435

6a indicates mountain glacier surfaces are incorrectly being captured in this region. This 436

poses a challenge to other tracking studies, and mountainous areas are sometimes 437

removed from analysis [Neu et al., 2013]. As a future improvement, we could develop a 438

dynamic threshold criteria rather than a fixed brightness temperature value. [Hennon et 439

al., 2011]. 440


Another challenge lies in the early termination of cloud clusters due to splits and mergers. 441

As clouds evolve, they continuously split and merge, each of which resets the lifetime 442

clock to zero. Only the largest, most well defined clusters avoid this in their lifetimes. 443

This is a limitation of this specific technique but the tracking algorithm could be refined 444

in the future to track features that do not have an easily defined shape, such as wintertime 445

midlatitudes storms or the movement of clouds that are part of atmospheric rivers. 446

In spite of such limitations, there are many promising areas of future work. The cluster 447

tracking provided in this study can be combined with other event based datasets, such as 448

the TRMM precipitation feature (TRMM-PF) dataset developed by Liu et al. [2008]. 449

TRMM-PF has been extensively used to study the scale and intensity of rainfall events 450

and can infer life cycle stage from the vertical profiles obtained from the precipitation 451

radar. By combing TRMM-PF with our IR-based cloud tracks, rain features can be 452

studied in context of their entire life cycle and trajectory, overcoming the sampling limits 453

of polar orbiting satellites, to further our understanding of precipitating cloud systems. 454

ACKNOWLEDGEMENTS 455

This research was supported by the NASA Earth System Data Records Uncertainty 456

Analysis Program and NASA’s Precipitation Measurement Missions (PMM) program. 457

Computing resources were provided by the NASA Center for Climate Simulation. The 458

data used in this study are available online through the Goddard Earth Sciences Data and 459

Information Services Center’s Mirador Search tool: http://mirador.gsfc.nasa.gov. Upon 460

publication of the manuscript, we plan to distribute cloud cluster tracks created in this 461

study. 462


REFERENCES 463

Adler, R. F. et al. (2003), The Version-2 Global Precipitation Climatology Project 464

(GPCP) Monthly Precipitation Analysis (1979–Present), J. Hydrometeor., 4(6), 465

1147–1167, doi:10.1175/1525-7541. 466

Boer, E. R., and V. Ramanathan (1997), Lagrangian approach for deriving cloud 467

characteristics from satellite observations and its implications to cloud 468

parameterization, J. Geophys. Res., 102(D17), 21383–21399, doi:10.1029/97JD00930. 469

Bengtsson, L., K. I. Hodges, and E. Roeckner (2006), Storm tracks and climate change, J. 470

Climate, 19(15), 3518–3543. 471

Bengtsson, L., K. I. Hodges, and N. Keenlyside (2009), Will Extratropical Storms 472

Intensify in a Warmer Climate?, J. Climate, 22(9), 2276–2301, 473

doi:10.1175/2008JCLI2678.1. 474

Blamey, R. C., and C. J. C. Reason (2012), Mesoscale Convective Complexes over 475

Southern Africa, J. Climate, 25(2), 753–766, doi:10.1175/JCLI-D-10-05013.1. 476

Carvalho, L. M. V., and C. Jones (2001), A Satellite Method to Identify Structural 477

Properties of Mesoscale Convective Systems Based on the Maximum Spatial 478

Correlation Tracking Technique (MASCOTTE), Journal of Appl. Meteorol, 40(10), 479

1683–1701, doi:10.1175/1520-0450. 480


Chen, S. S., R. A. Houze Jr., and B.E. Mapes (1996), Multiscale variability of deep 481

convection in relation to large-scale circulation in TOGA COARE, Journal of the 482

Atmospheric Sciences, 53(10), 1380. 483

Chen, S. S., and R. A. Houze (1997), Diurnal variation and life-cycle of deep convective 484

systems over the tropical pacific warm pool, Q.J.R. Meteorol. Soc., 123(538), 357–485

388, doi:10.1002/qj.49712353806. 486

Ebert, E. E., J. E. Janowiak, and C. Kidd (2007), Comparison of Near-Real-Time 487

Precipitation Estimates from Satellite Observations and Numerical Models, Bull. 488

Amer. Meteor. Soc., 88(1), 47–64, doi:10.1175/BAMS-88-1-47. 489

Fiolleau, T., and R. Roca (2013), An Algorithm for the Detection and Tracking of 490

Tropical Mesoscale Convective Systems Using Infrared Images From Geostationary 491

Satellite, IEEE T. Geosci. Remot, 51(7), 4302–4315, 492

doi:10.1109/TGRS.2012.2227762. 493

Futyan, J. M., and A. D. Del Genio (2007), Deep Convective System Evolution over 494

Africa and the Tropical Atlantic, J. Climate, 20(20), 5041–5060, 495

doi:10.1175/JCLI4297.1. 496

Hennon, C. C., C. N. Helms, K. R. Knapp, and A. R. Bowen (2011), An Objective 497

Algorithm for Detecting and Tracking Tropical Cloud Clusters: Implications for 498

Tropical Cyclogenesis Prediction, J. Atmos. Ocean. Tech., 28(8), 1007–1018, 499

doi:10.1175/2010JTECHA1522.1. 500


Hoskins, B. J., and K. I. Hodges (2002), New perspectives on the Northern Hemisphere 501

winter storm tracks, Atmos. Sci, 59(6), 1041–1061. 502

Houze, R. A., Jr., K. L. Rasmussen, M. D. Zuluaga, and S. R. Brodzik (2015), The 503

variable nature of convection in the tropics and subtropics: A legacy of 16 years of 504

the Tropical Rainfall Measuring Mission (TRMM) satellite. Rev. Geophys., 53, 505

doi:10.1002/2015RG000488. 506

Hou, A. Y., R. K. Kakar, S. Neeck, A. A. Azarbarzin, C. D. Kummerow, M. Kojima, R. 507

Oki, K. Nakamura, and T. Iguchi (2013), The Global Precipitation Measurement 508

Mission, Bull. Amer. Meteor. Soc., 95(5), 701–722, doi:10.1175/BAMS-D-13-509

00164.1. 510

Hsu, K., H. V. Gupta, X. Gao, and S. Sorooshian (1999), Estimation of physical variables 511

from multichannel remotely sensed imagery using a neural network: Application to 512

rainfall estimation, Water Resour. Res., 35(5), 1605–1618, 513

doi:10.1029/1999WR900032. 514

Huffman, G.J., D. Bolvin, D. Braithwaite, K. Hsu, R. Joyce, and P. Xie (2013). NASA 515

Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for 516

GPM (IMERG). Algorithm Theoretical Basis Doc., version 4.1, NASA, 29 pp. 517

[http://pmm.nasa.gov/sites/default/files/document_files/IMERG_ATBD_V4.1.pdf.] 518

Huffman, G. J., D. T. Bolvin, E. J. Nelkin, D. B. Wolff, R. F. Adler, G. Gu, Y. Hong, K. 519

P. Bowman, and E. F. Stocker (2007), The TRMM Multisatellite Precipitation 520


Analysis (TMPA): Quasi-global, multiyear, combined-sensor precipitation estimates 521

at fine scales, Journal of Hydrometeorology, 8(1), 38–55. 522

Janowiak, J. E., R. J. Joyce, and Y. Yarosh (2001), A Real–Time Global Half–Hourly 523

Pixel–Resolution Infrared Dataset and Its Applications, Bull. Amer. Meteor. Soc., 524

82(2), 205–217, doi:10.1175/1520-0477. 525

Joyce, R. J., J. E. Janowiak, P. A. Arkin, and P. Xie (2004), CMORPH: A method that 526

produces global precipitation estimates from passive microwave and infrared data at 527

high spatial and temporal resolution, Journal of Hydrometeorology, 5(3), 487–503. 528

Kikuchi, K., and B. Wang (2008), Diurnal Precipitation Regimes in the Global Tropics*, 529

Journal of Climate, 21(11), 2680–2696, doi:10.1175/2007JCLI2051.1. 530

Kubota, T., and S. Shige (2007), Global Precipitation Map Using Satellite-Borne 531

Microwave Radiometers by the GSMaP Project: Production and Validation, IEEE 532

Trans. Geosci. Remote Sens., 45(7), 2259 – 2275, doi:10.1109/TGRS.2007.895337. 533

Kummerow, C. et al. (2000), The Status of the Tropical Rainfall Measuring Mission 534

(TRMM) after Two Years in Orbit, J. Appl. Meteor., 39(12), 1965–1982, 535

doi:10.1175/1520-0450(2001)040<1965:TSOTTR>2.0.CO;2. 536

Kunkel, K. E., T. R. Karl, D. R. Easterling, K. Redmond, J. Young, X. Yin, and P. 537

Hennon (2013), Probable maximum precipitation and climate change, Geophys. Res. 538

Lett., 40(7), 1402–1408, doi:10.1002/grl.50334. 539


Laing, A. G., and J. Michael Fritsch (1997), The global population of mesoscale 540

convective complexes, Q.J.R. Meteorol. Soc., 123(538), 389–405, 541

doi:10.1002/qj.49712353807. 542

Lakshmanan, V., K. Hondl, and R. Rabin (2009), An Efficient, General-Purpose 543

Technique for Identifying Storm Cells in Geospatial Images, J. Atmos. Ocean. 544

Technol., 26(3), 523–537, doi:10.1175/2008JTECHA1153.1. 545

Li, J., K. Hsu, A. AghaKouchak, and S. Sorooshian (2015), An object-based approach for 546

verification of precipitation estimation, International Journal of Remote Sensing, 547

36(2), 513–529, doi:10.1080/01431161.2014.999170. 548

Liu, C., E. J. Zipser, D. J. Cecil, S. W. Nesbitt, and S. Sherwood (2008), A Cloud and 549

Precipitation Feature Database from Nine Years of TRMM Observations, J. Appl. 550

Meteor. Climatol., 47(10), 2712–2728, doi:10.1175/2008JAMC1890.1. 551

Liu, C., and E. J. Zipser (2009), “Warm Rain” in the Tropics: Seasonal and Regional 552

Distributions Based on 9 yr of TRMM Data, J. Climate, 22(3), 767–779, 553

doi:10.1175/2008JCLI2641.1. 554

Liu, C., and E. J. Zipser (2015), The global distribution of largest, deepest, and most 555

intense precipitation systems: largest, deepest and strongest storms, Geophys. Res. 556

Lett., 42(9), doi:10.1002/2015GL063776. 557

Lovejoy, S., and D. Schertzer (2006), Multifractals, cloud radiances and rain, J. Hydrol., 558

322(1-4), 59–88, doi:10.1016/j.jhydro1.2005.02.042. 559


Machado, L. A. T., M. Desbois, and J.-P. Duvel (1992), Structural Characteristics of 560

Deep Convective Systems over Tropical Africa and the Atlantic Ocean, Monthly 561

Weather Review, 120(3), 392–406, doi:10.1175/1520-562

0493(1992)120<0392:SCODCS>2.0.CO;2. 563

Machado, L. A. T., W. B. Rossow, R. L. Guedes, and A. W. Walker (1998), Life cycle 564

variations of mesoscale convective systems over the Americas, Mon. Weather Rev., 565

126(6), 1630–1654. 566

Machado, L. A. T., and H. Laurent (2004), The convective system area expansion over 567

Amazonia and its relationships with convective system life duration and high-level 568

wind divergence, Monthly weather review, 132(3), 714–725. 569

Maddox, R. A. (1980), Meoscale Convective Complexes, Bull. Amer. Meteor. Soc., 570

61(11), 1374–1387, doi:10.1175/1520-0477. 571

Mapes, B. E., and R. A. Houze (1993), Cloud Clusters and Superclusters over the 572

Oceanic Warm Pool, Mon. Weather Rev., 121(5), 1398–1416, doi:10.1175/1520-0493. 573

Morel, C., and S. Senesi (2002), A climatology of mesoscale convective systems over 574

Europe using satellite infrared imagery. II: Characteristics of European mesoscale 575

convective systems, Q.J.R. Meteorol. Soc., 128(584), 1973–1995, 576

doi:10.1256/003590002320603494. 577

Neu, U. et al. (2013), IMILAST: A Community Effort to Intercompare Extratropical 578

Cyclone Detection and Tracking Algorithms, Bull. Amer. Meteor. Soc., 94(4), 529–579

547, doi:10.1175/BAMS-D-11-00154.1. 580


Parzen, E. (1962), On Estimation of a Probability Density Function and Mode, Ann. Math. 581

Statist., 33(3), 1065–1076, doi:10.1214/aoms/1177704472. 582

Rosenblatt, M. (1956), Remarks on Some Nonparametric Estimates of a Density Function, 583

Ann. Math. Statist., 27(3), 832–837, doi:10.1214/aoms/1177728190. 584

Sinclair, M. R. (1994), An Objective Cyclone Climatology for the Southern Hemisphere, 585

Mon. Wea. Rev., 122(10), 2239–2256, doi:10.1175/1520-0493. 586

Stephens, G. L., T. L’Ecuyer, R. Forbes, A. Gettlemen, J.-C. Golaz, A. Bodas-Salcedo, K. 587

Suzuki, P. Gabriel, and J. Haynes (2010), Dreary state of precipitation in global 588

models, J. Geophys. Res., 115, D24211, doi:10.1029/2010JD014532. 589

Tadesse, A., and E. N. Anagnostou (2009), The Effect of Storm Life Cycle on Satellite 590

Rainfall Estimation Error, J. Atmos. Ocean., 26(4), 769–777, 591

doi:10.1175/2008JTECHA1129.1. 592

Teng, H.-F., C.-S. Lee, and H.-H. Hsu (2014), Influence of ENSO on formation of 593

tropical cloud clusters and their development into tropical cyclones in the western 594

North Pacific: Influence of ENSO on TCC Formation, Geophysical Research Letters, 595

41(24), 9120–9126, doi:10.1002/2014GL061823. 596

Thorncroft, C., and K. Hodges (2001), African Easterly Wave Variability and Its 597

Relationship to Atlantic Tropical Cyclone Activity, J. Climate, 14(6), 1166–1179, 598

doi:10.1175/1520-0442(2001)014<1166:AEWVAI>2.0.CO;2. 599


Velasco, I., and J. Fritsch (1987), Mesoscale Convective Complexes in the America, J. 600

Geophys. Res.-Atmos., 92(D8), 9591–9613, doi:10.1029/JD092iD08p09591. 601

Vila, D. A., L. A. T. Machado, H. Laurent, and I. Velasco (2008), Forecast and Tracking 602

the Evolution of Cloud Clusters (ForTraCC) using satellite infrared imagery: 603

Methodology and validation, Wea. Forecasting, 23(2), 233–245. 604

Westra, S., H. J. Fowler, J. P. Evans, L. V. Alexander, P. Berg, F. Johnson, E. J. Kendon, 605

G. Lenderink, and N. M. Roberts (2014), Future changes to the intensity and 606

frequency of short-duration extreme rainfall, Rev. Geophys., 52(3), 2014RG000464, 607

doi:10.1002/2014RG000464. 608

Wilks, D. S. (2011), Statistical methods in the atmospheric sciences, International 609

geophysics series v. 100, 3rd ed., Elsevier/Academic Press, Amsterdam ; Boston. 610

Williams, M., and R. A. Houze (1987), Satellite-Observed Characteristics of Winter 611

Monsoon Cloud Clusters, Mon. Weather Rev., 115(2), 505–519, doi:10.1175/1520-612

0493. 613

Xie, P., and A.-Y. Xiong (2011), A conceptual model for constructing high-resolution 614

gauge-satellite merged precipitation analyses, J. Geophys. Res., 116(D21), D21106, 615

doi:10.1029/2011JD016118. 616

Zahraei, A., K. Hsu, S. Sorooshian, J. J. Gourley, Y. Hong, and A. Behrangi (2013), 617

Short-term quantitative precipitation forecasting using an object-based approach, J. 618

Hydrol., 483, 1–15, doi:10.1016/j.jhydrol.2012.09.052. 619


Zipser, E. J., D. J. Cecil, C. Liu, S. W. Nesbitt, and D. P. Yorty (2006), Where are the 620

most intense thunderstorms on earth?, Bull. Amer. Meteorol. Soc., 87(8), 1057, 621

doi:10.1175/BAMS-87-8-1057. 622

Zolina, O., and S. K. Gulev (2002), Improving the Accuracy of Mapping Cyclone 623

Numbers and Frequencies, Mon. Wea. Rev., 130(3), 748–759, doi:10.1175/1520-0493. 624

625

1

Figure 1. (a) Globally merged map of IR brightness temperature from NCEP/CPC Cloud brightness temperature dataset for 23:00 GMT June 28, 2012. (b) Cloud clusters captured by ForTraCC after applying temperature and size thresholds. Shading represents cloud brightness temperature.

b)

K

a)

K

2

Figure 2. Schematic of area-overlap handling of continuous systems (c), merging systems (m), and splitting systems (s). The image was taken of thunderstorms developing over the American Midwest beginning at 3:00pm EST on June 30, 2012. Yellow represents the initial time, orange 1.5 hours later, and red 3.0 hours after initial detection.

Figure 3. Cloud cluster tracks from Dec 1-4, 2001 produced from using the ForTrACC algorithm. A few days of tracking yield a large number of clusters and their movement begins to trace out large-scale atmospheric patterns.

3

Figure 4. Climatology of cloud cluster trajectories in (a) DJF and (b) JJA, 2002-2012 with 6-9 hour lifetimes binned by 2° x 2°. Lines show average displacement of all cloud clusters that initiated at the same point over the 11-year period studied. Coloring indicates net zonal movement of clusters. Grid boxes with fewer than five initiations were not displayed.

b)

a)

4

Figure 5. (a) Number of detected events globally by lifetime and size for the entire record. (b) The kernel density estimate of cloud clusters at their maximum areal extent for each lifetime group over the entire study period (DJF and JJA, 2002-2012).

b)

a)

5

Figure 6. Mean seasonal frequency of clusters for (a) DJF and (b) JJA at their maximum areal extent. The figure shows the average seasonal count of events over the study period, binned by 2° x 2° latitude-longitude. Warmer colors represent higher counts while cooler colors represent fewer observations. White grid boxes have ten or fewer cloud clusters across the 11 year period.

b)

a)

6

Figure 7. The composite of the 11-year DJF mean annual frequency of cloud cluster overpasses for El Niño and La Niña, binned by 2° x 2° latitude-longitude, at their maximum areal extent. Warm or cool event years were selected based on the NINO3.4 sea surface temperature anomaly index. White grid boxes have three or fewer cloud clusters.

7

Figure 8. The global average life cycle evolution for new cloud clusters with varying life times. Each curve represents the average properties of millions of clusters grouped by life span. The shortest lines are short-lived events longer lines are long-lived. Curves show how the size changes as a function of the clusters’ lifetime. Dashed curve is a regression fitted to the maximum of each curve. Seasons are defined by the ±25° latitude line.

c)

e)

a)

d)

f)

b)

8

Figure 9. Same as Figure 8, but showing how the average brightness temperature changes as a function of time and lifetime.

c)

e)

a)

d)

f)

b)

9

Figure 10. Total instantaneous rainfall contribution as a function of rain rate captured by clusters in the (a, b) Northern Hemisphere, (c, d) tropics, and (e, f) Southern Hemisphere in June and December 2014. The distribution is based on coincident cloud clusters and passive microwave-based rainfall estimates from the IMERG dataset.

c)

e)

a)

d)

f)

b)

10

Figure 11. Diurnal variation in local solar time (LST) cloud cluster initiation for (a) DJF and (b) JJA, binned by 2° x 2° latitude-longitude. Each box shows the timing of maximum occurrence of cluster formation for the 11-year record. Only clusters with a lifetime greater than two hours are included.

11

Figure 12. The kernel density of local solar time of the life cycle stage in two regions, 0°-30°N and 60°-90°E (South Asia) and 0-40°N, 50°W-20°E (West Africa).

Figure 13. The kernel density of local solar time of initiation in three-hourly lifetime bins for two regions, 0°-30°N and 60°-90°E (South Asia) and 0-40°N, 50°W-20°E (West Africa).

c)

a)

d)

b)

c)

a)

d)

b)

Documents

A Lagrangian analysis of cold cloud clusters and …...Esmaili et al., 2016 1 1 A Lagrangian analysis of cold cloud clusters and their life cycles with 2 satellite observations 3 4